Start Date
11-12-2016 12:00 AM
Description
Medication prescription rankings and demands prediction could benefit both medication consumers and pharmaceutical companies from various aspects. Our study predicts the medication prescription rankings focusing on patients’ medication switch and combination behavior, which is an innovative genre of medication knowledge that could be learned from unstructured patient generated contents. We first construct two supervised machine learning systems for medication references identification and medication relations classification from unstructured patient’s reviews. We further map the medication switch and combination relations into directed and undirected networks respectively. An adjusted transition in and out (ATIO) system is proposed for medication prescription rankings prediction. The proposed system demonstrates the highest positive correlation with actual medication prescription amounts comparing to other network-based measures. In order to predict the prescription demand changes, we compare four predictive regression models. The model incorporated the network-based measure from ATIO system achieve the lowest mean square errors.
Recommended Citation
Gao, Yuanyuan and Hu, Paul J., "Predicting Medication Prescription Rankings with Medication Relation Network" (2016). ICIS 2016 Proceedings. 10.
https://aisel.aisnet.org/icis2016/ISHealthcare/Presentations/10
Predicting Medication Prescription Rankings with Medication Relation Network
Medication prescription rankings and demands prediction could benefit both medication consumers and pharmaceutical companies from various aspects. Our study predicts the medication prescription rankings focusing on patients’ medication switch and combination behavior, which is an innovative genre of medication knowledge that could be learned from unstructured patient generated contents. We first construct two supervised machine learning systems for medication references identification and medication relations classification from unstructured patient’s reviews. We further map the medication switch and combination relations into directed and undirected networks respectively. An adjusted transition in and out (ATIO) system is proposed for medication prescription rankings prediction. The proposed system demonstrates the highest positive correlation with actual medication prescription amounts comparing to other network-based measures. In order to predict the prescription demand changes, we compare four predictive regression models. The model incorporated the network-based measure from ATIO system achieve the lowest mean square errors.